AI for Automatic Defects Detection on Large Civil Works: Our Adventure from the Database Creation to a First Efficient Pipeline
Year: 2025
Type: conference-paper
Abstract: EDF, as a responsible operator, inspects and maintains the civil engineering structures of its nuclear power plants, which often exceed several tens of thousands of square meters in surface area. These inspections are carried out using drones, generating orthophotos of the surfaces. Manual analysis of these images to identify defects such as cracks, corroded bars, and spalling is costly and time-consuming. To address these issues, EDF is developing an AI-based tool for automated analysis to detect, locate, and measure these defects. The project includes a state-of-the-art review, creation of a training database, image preprocessing, data augmentation, neural network implementation, performance metrics selection, and optimization of the processing chain. The first results obtained and feedback on this image processing pipeline will be presented.
Authors: PIERRE STEPHAN, PAULINE LAVIRON-PETIT, SELIM BEN-ABDALLAH, THIERRY LACOUR, GUSTAVO MEDINA VASQUEZ, QIUFENG ZHENG
Institutions: Électricité de France (France), EDF Energy (United Kingdom)
Language: English
Subfield: Civil and Structural Engineering
Field: Engineering
Domain: Physical Sciences
Open Access status: bronze
OpenAlex ID: W4415216826
FWCI:
Cites:
Cited by:
Related to: